An interesting background for this topic is The (Mis)Behaviour of Markets: A Fractal View of Risk, Ruin and Reward" by Benoit B. Mandelbrot. TLDR - there are patterns, but they are not truly predictive of market behaviour, it just looks like it.
Read it a few times. Then ran independent research — not to disagree, but to find where intuition ends and fact begins.
The facts came out stronger.
2025-2026 produced a convergence that's hard to ignore: five independent fields — neuroscience, financial ML, Chinese systemic risk models, network physics, LLM evaluation — independently arrived at the same architecture. The regime detector sits before the model, not inside it. PNAS October 2025 via Kuramoto showed that proximity to a critical transition predicts collapse trajectory — something no return-based model can structurally see.
Millennium and Point72 each lost $1.5 billion not because the models were bad. They fell along a trajectory physics already knew.
What's striking is that neuroscience showed the same thing from the inside: the locus coeruleus isn't part of the predictor — it's a separate system above it that decides when the brain's model is no longer valid and triggers a rewrite.
I'm building exactly this detector — through cross-asset synchronization, not price prediction. A system that knows when to stop matters more than one that always answers.
Your article frames this precisely: the problem isn't technical — it's epistemological. The model doesn't know what it doesn't know.
One question after reading: do you see a difference between how American and Chinese approaches behave at the moment of transition? The data suggests there is one — and it's architectural.
I agree, there’s a very big difference between true predictive power and the ability to predict retrospectively. I don’t think the study I read was tested over several market crashes, and on anything beyond a stock index e.g currency market or credit markets. It is definitely to be taken with a grain of salt.
An interesting background for this topic is The (Mis)Behaviour of Markets: A Fractal View of Risk, Ruin and Reward" by Benoit B. Mandelbrot. TLDR - there are patterns, but they are not truly predictive of market behaviour, it just looks like it.
Sounds interesting, will definitely look it up!
Read it a few times. Then ran independent research — not to disagree, but to find where intuition ends and fact begins.
The facts came out stronger.
2025-2026 produced a convergence that's hard to ignore: five independent fields — neuroscience, financial ML, Chinese systemic risk models, network physics, LLM evaluation — independently arrived at the same architecture. The regime detector sits before the model, not inside it. PNAS October 2025 via Kuramoto showed that proximity to a critical transition predicts collapse trajectory — something no return-based model can structurally see.
Millennium and Point72 each lost $1.5 billion not because the models were bad. They fell along a trajectory physics already knew.
What's striking is that neuroscience showed the same thing from the inside: the locus coeruleus isn't part of the predictor — it's a separate system above it that decides when the brain's model is no longer valid and triggers a rewrite.
I'm building exactly this detector — through cross-asset synchronization, not price prediction. A system that knows when to stop matters more than one that always answers.
Your article frames this precisely: the problem isn't technical — it's epistemological. The model doesn't know what it doesn't know.
One question after reading: do you see a difference between how American and Chinese approaches behave at the moment of transition? The data suggests there is one — and it's architectural.
I agree, there’s a very big difference between true predictive power and the ability to predict retrospectively. I don’t think the study I read was tested over several market crashes, and on anything beyond a stock index e.g currency market or credit markets. It is definitely to be taken with a grain of salt.